> What leaves me even more confused is that the entire log message > does not make it clear what the end-user observable problem the > patch is trying to solve. > > Is this "we sometimes follow and sometimes fail to follow refs while > fetching"? Does it affect all protocol versions and transports, or > only just selected few (and if so which ones)?
Normally I would respond by creating a new patch with the answer in its commit message, but I'm now not sure about whether it's better to revert back to the non-"fetched_refs" API entirely (as I explained in the reply to Peff I just sent [1]), so I'll answer your questions here for now: - Yes. We fail to follow when we fetch at least one ref that is up-to-date and one ref that is not, and when we're using the "fetch" command in a remote helper (for example, HTTP protocol v0). - I haven't checked exhaustively, but as far as I know, affects HTTP protocol v0, and does not affect anything using connect or stateless-connect (e.g. HTTP protocol v2, ssh). When I create a new patch, I'll also include these answers in its commit message. [1] https://public-inbox.org/git/20180731232343.184463-1-jonathanta...@google.com/
